The fastest way to build AI apps

We’re excited to introduce Writer AI Studio, the fastest way to build AI apps, products, and features. Writer’s unique full-stack design makes it easy to prototype, deploy, and test AI apps – allowing developers to build with APIs, a drag-and-drop open-source Python framework, or a no-code builder, so you have flexibility to build the way you want.

Writer comes with a suite of top-ranking LLMs and has built-in RAG for easy integration with your data. Check it out if you’re looking to streamline how you build and integrate AI apps.

Start building Writer AI Studio.

Today:

• Microsoft Autonomous Agents

• IBM Introduces Granite 3.0 AI

• xAI API Launches for Developers

• ByteDance Dismisses Intern Over AI

• News Corp Sues Perplexity AI

Microsoft Autonomous Agents

Microsoft is enhancing its AI tools by introducing autonomous agents—software programs that perform tasks without constant human guidance—in Copilot Studio and Dynamics 365. Starting next month, Copilot Studio will let users create and manage these agents. Ten new agents will assist teams in sales, service, finance, and supply chain management. 

For example, a sales agent helps prioritize leads and personalize outreach, while a supply chain agent monitors suppliers to prevent delays. Companies like McKinsey and Pets at Home are using these agents to save time and reduce costs. Microsoft has also improved productivity by using Copilot and agents internally.

IBM Introduces Granite 3.0 AI

IBM introduced Granite 3.0, a new suite of AI models optimized for business use, during its annual TechXchange event. Released under the open-source Apache 2.0 license, Granite 3.0 offers high performance, safety features, and efficiency for various enterprise tasks. The models include general-purpose, safety-focused, and Mixture-of-Experts versions, suitable for CPU-based deployments. 

IBM also launched new tools in watsonx.ai, supporting AI deployment, coding assistance, and autonomous agents. The Granite models will enhance IBM's consulting platform, empowering its 160,000 consultants to deliver faster solutions for clients. Granite 3.0 is available for commercial use on watsonx and other platforms.

xAI API Launches for Developers

Elon Musk's xAI has launched a new API that allows third-party developers to build applications using its Grok large language models. The API provides access to Grok-2 and Grok-2 mini models, supporting text, code, image generation, and function calling for real-world tasks like IoT integration. It offers a user-friendly web console with features like usage tracking, team management, and enhanced security. 

Although the xAI API is currently in beta, developers can sign up and begin exploring its capabilities. Pricing is slightly higher than competitors like OpenAI, with a range of new features aimed at simplifying enterprise adoption.

ByteDance Dismisses Intern Over AI

ByteDance, the owner of TikTok, dismissed an intern in August 2024 for allegedly sabotaging an artificial intelligence (AI) project. The company stated that the intern “maliciously interfered” with the training of AI models used in a research project, but clarified that its official AI products and large language models were not affected. 

The incident followed rumors on Chinese social media, which ByteDance dismissed as exaggerated, including claims of significant disruptions. ByteDance has informed the intern’s university and relevant industry associations about the misconduct. This comes amid heightened scrutiny of generative AI models and ByteDance’s ongoing challenges in the U.S.

News Corp Sues Perplexity AI

The Wall Street Journal and New York Post, both owned by News Corp, have filed a lawsuit against AI startup Perplexity, accusing it of copyright infringement. The lawsuit claims that Perplexity is copying their copyrighted news content and using it to generate responses to user queries, thus diverting traffic that would otherwise go to the publishers' websites. 

This legal action represents the latest in a series of disputes between news organizations and AI companies over the unauthorized use of content. News Corp is seeking to block Perplexity from using its material and recover damages.

🧠RESEARCH

Sabotage Evaluations for Frontier Models

The paper explores the risk of advanced AI models sabotaging human oversight and decision-making, especially in monitoring their behavior and deployment. It introduces threat models and evaluations to assess if AI models could sabotage organizations' activities. Testing on Claude 3 models shows that current mitigations are effective, but stronger measures may be needed as AI capabilities grow. The paper emphasizes the importance of mitigation-aware evaluations and simulating large-scale risks using smaller-scale tests.

UCFE: A User-Centric Financial Expertise Benchmark for Large Language Models

UCFE, a new framework for evaluating large language models in real-world financial tasks. It combines expert evaluations with dynamic user interactions. A study with 804 participants shaped the dataset, and 12 LLM services were tested. Results show strong alignment with human preferences, validating UCFE’s effectiveness in assessing LLMs.

Mini-Omni2: Towards Open-source GPT-4o with Vision, Speech and Duplex Capabilities

Mini-Omni2, an open-source model designed to replicate GPT-4o’s multi-modal capabilities, including understanding and responding to visual, auditory, and text inputs. Mini-Omni2 integrates pretrained encoders and uses a three-stage training process to align modalities. It enables real-time, flexible interaction, offering insights for further multi-modal model development.

FiTv2: Scalable and Improved Flexible Vision Transformer for Diffusion Model

FiTv2, an upgraded Flexible Vision Transformer (FiT) designed to generate images with any resolution and aspect ratio. By treating images as token sequences instead of fixed grids, FiTv2 improves resolution flexibility and eliminates biases from cropping. It features faster convergence, enhanced adaptability, and scalability for high-resolution image generation. All codes and models are available for further research.

DreamVideo-2: Zero-Shot Subject-Driven Video Customization with Precise Motion Control

DreamVideo-2, a zero-shot video customization framework that generates videos with specific subjects and motion trajectories using a single image and bounding box sequence, without test-time fine-tuning. It employs reference attention and a mask-guided motion module for precise control, balancing subject learning and motion. DreamVideo-2 surpasses current methods in customization and motion control, with plans to release the dataset, code, and models publicly.

🛠️TOP TOOLS

AI Desk - AI-powered Customer Service on your website

All Hands - Open Source Agents for Developers

Strella - AI-moderated interviews and instant synthesis, powering smarter & faster decisions

BrowserCopilot - AI browser copilot.

Wordtune - AI writer that can paraphrase, rewrite, correct your grammar, and more.

📲SOCIAL MEDIA

🗞️MORE NEWS

• Apple CEO Tim Cook defended the company’s delayed entry into AI, emphasizing their focus on delivering quality products over being first. He highlighted Apple’s philosophy: "Not first, but best."

• Researchers at UCLA developed an AI model, SLIViT, that quickly analyzes 3D medical images for disease biomarkers, outperforming existing models. It uses 2D pre-training and fine-tuning on smaller 3D datasets, making expert-level analysis scalable and cost-efficient.

• Lumen Technologies and Meta have partnered to expand Meta’s AI network capacity, enhancing its infrastructure for advanced AI tasks. Lumen provides secure, flexible bandwidth, supporting Meta’s complex computing needs for a more connected, AI-driven future.

• Nvidia is partnering with Aidoc to create the BRIDGE framework, which aims to guide AI adoption in healthcare by offering structured integration and deployment guidelines. This collaboration strengthens Nvidia's role in advancing healthcare AI solutions.

• Honeywell has partnered with Google to bring its Gemini AI to the industrial sector. This collaboration will enhance maintenance efficiency, boost productivity, and help address labor shortages by integrating AI into Honeywell’s Forge IoT platform, with deployment starting in 2025.